Log-based Collaborative Automatic Image Annotation

In recent years, science technology has maintained the momentum of rapiddevelopment, which the development of the Internet and multimedia technology isparticularly prominent, and a variety of high-definition digital multimedia devices arealso very popular, then produce massive image, how to effectively deal with theexplosive growth image and how to quickly retrieve the target image in the mass imagehave become a hot issue, Automatic image annotation technology can improve retrievalefficiency therefore its attention has been paid. The main research contents are asfollows.First, for the problem that the BOW model lacks the spatial relationship betweenlocal visual features when we use it in the image classification and identification, thispaper proposes bag-of-phrases to expansion the original visual dictionary, it means thesystem dig out those visual vocabulary which often occurs together in the close spatiallocation, then make up this deficiency. However, when we use BOW model to representan image, it still exists the problem of local visual features disorder, so this paperproposes based on the BOW’s Spatial pyramid model, it uses Spatial pyramid model tohierarchical dividing the image in the image space instead of feature space, and thenrepresent the image based on visual dictionary, in this way, it Alleviate the problem offeature disorder by partial.Secondly, on the basis of analysis many annotation model and also combined withuser feedback logs and collaborative idea, this paper proposes Log-based collaborativeautomatic image annotation algorithm. we use incremental association rule to mine logmessages according to its characteristics to remove the noisy data and expand theamount of image annotation words with the thought of collaborative filtering. Then, useWordNet to constructing semantic relationships between the various annotationthesaurus for further improve the relationship between each image’s annotation.Finally, under the framework of HPM, we use the based on the BOW’s Spatial pyramidmodel to compute the image underlying feature, and then combining the relationshipbetween the annotation to improve the performance of image annotation.Finally, this paper integrates the functional modules and algorithms of annotationsystem, and then programming an automatic image annotation system. This system cannot only complete the image retrieval, but also collect and treat user feedback log and achieve subsequent annotation algorithm.